On 2026/02/09 03:28, Kurt Mosiejczuk wrote: > On Sat, Feb 07, 2026 at 10:53:06AM +0000, Stuart Henderson wrote: > > > Build failures: 46 > > > I looked at some of these, obviously some are ports which don't work on > > sparc64, but there are some which look like machine or OS problems, e.g. > > > > https://cranky.work/sparc64/2026-02-04/math/lapack.log > > > build stopped strangely, then when restarted with the same build dir > > a .d file is corrupt > > I had two nodes kernel panic. This was going on one of the two nodes.
I don't think build dir contents can be trusted after a crash. FWIW, on the i386 builders I mount /pobj with "...,noauto 0 0" and do this at boot: : i386.p; head -3 /etc/rc.local newfs /dev/rsd0m mount /pobj chown _pbuild:_pbuild /pobj (also saves time fsck'ing). > > config.status: creating ntpdate/Makefile > > config.status: creating ntpdc/Makefile > > awk: trying to access out of range field -2147483648 > > input record number 1, file > > source line number 316 > > config.status: error: could not create ntpdc/Makefile > > Yeah. This is the classic sparc64 error I've been experiencing for years now. > It also takes the form of mentions of NaN or perl crashing at a line of "Use > X" > where X is a normal perl version but has been munged into something even > perl doesn't want to deal with. ah, I was aware of the perl one but not awk. I wonder what's going on here. > > config.status: creating libvslvm.pc > > awk: trying to access out of range field -2147483648 > > input record number 1, file > > source line number 364 > > config.status: error: could not create libvslvm.pc > > > Another error I'm betting comes from the same root cause is > > "panic: Unexpectedly didn't find a dot radix character in '-nan' at" some > perl file > > --Kurt Does sound plausible. I've not seen that in any i386 build failures.
